Fact · Dictation on Mac
3times faster, at least (more than x3)
The whisper.cpp project says that running the Whisper encoder on the Apple Neural Engine through Core ML can be more than three times faster than CPU-only execution on Apple silicon.
Checked against its source on October 16, 2026
In the source's own words
“This can result in significant speed-up - more than x3 faster compared with CPU-only execution”
whisper.cpp README, Georgi Gerganov, GitHub, 2026, Core ML support section.
Open the original- What it measures
- This applies to the encoder stage only, and requires a Core ML model to be generated and built into whisper.cpp. Ordinary users of a finished Mac app rarely see this setup.
- What it does not show
- The project does not say which Mac or which model size the figure refers to. Treat it as a general claim, not a benchmark.